A single-step diffusion-based MRI image super-resolution method and system

Abstract

The application discloses a kind of based on single-step diffusion's MRI image super-resolution method and system, it is related to image super-resolution technical field, method includes the following steps: constructing the conditional diffusion model based on deterministic sampling as teacher network;Student network based on distillation loss and consistency loss is constructed based on teacher network;MRI image super-resolution is realized using student network.The application reduces redundant steps in diffusion process layer by layer, concentrates the latent distribution of diffusion model to single-step generation, while fully retains the reconstruction ability of model to key details (such as tissue texture, structure boundary) of MRI image and image quality advantage.Compared with prior art MRI image super-resolution fast solution method, reasoning efficiency is significantly improved, and real-time demand under medical scene is satisfied;Moreover, in the quality method of generated high-resolution MRI image, it is flat with traditional method even has some surpass, provides more accurate image support for subsequent clinical application.

Description

Technical Field

[0001] This invention relates to the field of image super-resolution technology, and in particular to a method and system for MRI image super-resolution based on single-step diffusion. Background Technology

[0002] MRI (Magnetic Resonance Imaging) image super-resolution (SR) is an important research direction in the field of medical image processing. It aims to reconstruct high-resolution (HR) images from low-resolution (LR) MRI images using algorithms, thereby improving image detail clarity, spatial resolution, and diagnostic accuracy. Diffusion-based image super-resolution is a commonly used method; however, current diffusion-based methods generally rely on multi-step sampling processes, recovering high-resolution images from low-resolution images through step-by-step generation. While this step-by-step sampling strategy can fully utilize the generative capabilities of the diffusion model to accurately model complex image distributions, it also incurs significant computational costs and low inference efficiency, making it difficult to meet the real-time requirements of practical medical imaging applications. Chinese invention patent application CN120198291A discloses a generative image super-resolution method based on a diffusion model. The ResShift model, a typical representative of traditional conditional diffusion models, employs a core random multi-step sampling mechanism, requiring T iterations to obtain the restored result. This results in low inference efficiency and cannot meet the real-time requirements of MRI image super-resolution in medical scenarios. Furthermore, random sampling leads to uncertainty in the generated results, poor output stability, and multiple samplings may yield different results, making it difficult to guarantee the consistency and reliability of restoring key details of MRI images (such as tissue texture and structural boundaries). Additionally, it does not address the "multi-step redundancy" problem, resulting in high computational costs. To improve inference efficiency, some existing studies have proposed fast solvers to accelerate the diffusion process and reduce sampling steps. However, while these methods can reduce the number of samplings, they still require multiple iterations, failing to truly achieve one-step generation. Moreover, while accelerating the process, some image quality may be sacrificed. Summary of the Invention

[0003] The purpose of this invention is to solve the problems in the prior art.

[0004] In a first aspect, the present invention provides an MRI image super-resolution method based on single-step diffusion, comprising the following steps:

[0005] A conditional diffusion model based on deterministic sampling is constructed as the teacher network;

[0006] Construct a student network based on distillation loss and consistency loss using a teacher network;

[0007] Super-resolution of MRI images using student networks;

[0008] Specifically, the construction of a conditional diffusion model based on deterministic sampling as a teacher network involves using a conditional diffusion model based on random sampling as the base model, replacing the random sampling in its generation process with deterministic sampling, and thus obtaining the teacher network.

[0009] Preferably, the construction of the teacher network based on deterministic sampling includes the following steps:

[0010] A pre-trained conditional diffusion model, ResShift, is introduced, which uses random sampling from the initial state. Sampling yields the restored image ;

[0011] By modifying the random sampling of ResShift, a teacher network based on deterministic sampling is obtained; the deterministic sampling process is represented as follows:

[0012] ;

[0013] ;

[0014] in, This indicates a random sampling process. This represents the state at step t. Represents low-resolution MRI images. This represents a hyperparameter that monotonically increases with the time step. It is a unit impulse function; It is the restored image predicted from the pre-trained random ResShift; where Indicates the noise correction term. The offset parameter represents the deterministic sampling. This represents the weighting coefficients that change with the time step.

[0015] Preferably, the construction of the student network based on distillation loss and consistency loss includes the following steps:

[0016] The Unet network architecture is used as the foundation network for the student network;

[0017] Constructing distillation loss Constrained initial state Single-step output in a student network Compared with deterministic output The degree of approximation between them is expressed as:

[0018] ;

[0019] in, It is a student network that can directly predict the restored image in just one step; It is a deterministic single-step generating function of ResShift; This represents the mean squared error loss function;

[0020] Constructing realistic images Consistency loss , is represented as:

[0021] ;

[0022] The overall training objective for constructing the student network is expressed as:

[0023] ;

[0024] in, Indicates the training objective. This indicates finding the parameter corresponding to the minimum value. Represents the expectation operator based on multiple variables;

[0025] The basic network of the student network is trained using the overall training objective, resulting in a student network based on distillation loss and consistency loss.

[0026] Preferably, the student network includes:

[0027] The feature compression stage includes a shallow compression module, an intermediate compression module, and a deep compression module connected in sequence. These modules extract shallow, intermediate, and deep compressed feature maps of the input noisy image through convolution and downsampling, respectively.

[0028] The feature reconstruction stage includes a deep reconstruction module, an intermediate reconstruction module, and a shallow reconstruction module connected in sequence. The deep reconstruction module receives the deep compressed feature map and obtains the deep reconstructed feature map using convolution and normalization layers. The intermediate reconstruction module receives the deep reconstructed feature map and the deep compressed feature map transmitted through the feature fusion path, and obtains the intermediate reconstructed feature map using convolution and upsampling. The shallow reconstruction module receives the intermediate reconstructed feature map and the intermediate compressed feature map transmitted through the feature fusion path, and obtains the shallow reconstructed feature map using convolution and upsampling.

[0029] In the final output stage, the system receives the shallow reconstructed feature map and the shallow compressed feature map transmitted through the feature fusion path, and uses convolution and upsampling to generate the restored image at step T. .

[0030] Preferably, the shallow compression module, the intermediate compression module, and the deep compression module all include a convolutional layer, a cross-attention layer, and a downsampling unit with a stride of 2 connected in sequence.

[0031] The deep reconstruction module includes convolutional units, cross-attention layers, and normalization layers. The intermediate and shallow reconstruction modules each include convolutional layers, cross-attention layers, and upsampling units with a stride of 2 connected in sequence.

[0032] The final output stage consists of sequentially connected convolutional layers, cross-attention, and upsampling units with a stride of 2.

[0033] In a second aspect, the present invention also provides a single-step diffusion-based MRI image super-resolution system for implementing the single-step diffusion-based MRI image super-resolution method as described in any one of the first aspects, comprising:

[0034] The teacher network construction module builds a conditional diffusion model based on deterministic sampling as the teacher network;

[0035] The student network construction module builds a student network based on distillation loss and consistency loss, building upon the teacher network.

[0036] The image super-resolution module utilizes a student network to achieve MRI image super-resolution.

[0037] Thirdly, the present invention provides an electronic device, comprising:

[0038] One or more processors;

[0039] Storage device for storing one or more programs;

[0040] When the one or more programs are executed by the one or more processors, the one or more processors implement the single-step diffusion-based MRI image super-resolution method as described in any one of the first aspects.

[0041] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the single-step diffusion-based MRI image super-resolution method as described in any one of the first aspects.

[0042] Fifthly, the present invention provides a computer program product comprising a computer program that, when executed by a processor, implements the single-step diffusion-based MRI image super-resolution method as described in any one of the first aspects.

[0043] The present invention has the following beneficial effects: it adopts a dual-loss design, combining deterministic sampling with distillation loss. This makes the generated results more consistent, avoiding the uncertainty and fluctuations in generation quality caused by random sampling; a consistency loss is added. This ensures the similarity between the generated images and the real images, further improving the expressive power and generation accuracy of the student network. The dual loss design enables the model not only to mimic the performance of the teacher network, but also to better learn the feature distribution of the target image, thus making it more stable under different data distributions.

[0044] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments, but the present invention is not limited to the embodiments. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the steps of a single-step diffusion-based MRI image super-resolution method according to an embodiment of the present invention.

[0046] Figure 2 This is a flowchart illustrating an MRI image super-resolution method based on single-step diffusion according to an embodiment of the present invention.

[0047] Figure 3 This is a schematic diagram of the Unet module involved in a student network for a single-step diffusion-based MRI image super-resolution method according to an embodiment of the present invention.

[0048] Figure 4 This is a system structure diagram of an MRI image super-resolution system based on single-step diffusion, according to an embodiment of the present invention.

[0049] Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0050] See Figure 1 The diagram shows the method steps of an embodiment of the present invention, including the following steps:

[0051] S101, Construct a conditional diffusion model based on deterministic sampling as the teacher network;

[0052] S102, Construct a student network based on distillation loss and consistency loss based on the teacher network;

[0053] S103 utilizes a student network to achieve super-resolution of MRI images.

[0054] Specifically, to verify the model's performance in the low-resolution MRI image super-resolution task, experiments were conducted on the MRI dataset, with a subset of images retained as a validation set for model performance evaluation. Blurred, format-abnormal, and highly similar problematic samples were removed from the dataset. Furthermore, data augmentation operations such as rotation, flipping, cropping, and illumination adjustment were used to further enrich the data diversity. A bicubic filter degradation operator was used to perform degradation operations on the MRI images in the MRI dataset to adapt them for the image super-resolution (SR) task. To comprehensively evaluate the reconstruction effect of SR images, several quality evaluation metrics were introduced, including perceptual and non-perceptual quality metrics: Peak Signal-to-Noise Ratio (PSNR↑), Structural Similarity Index Measure (SSIM↑), and Learned Perceptual Image Patch Similarity (LPIPS↓).

[0055] Specifically, in step S101, in order to construct a teacher model based on deterministic sampling, the conditional diffusion model ResShift, pre-trained on the MRI dataset, is used as the base model of the teacher network, which includes the following steps:

[0056] S201 introduces a pre-trained conditional diffusion model, ResShift. Specifically, ResShift initially diffuses the low-resolution image... With real images The following forward noise addition process is used to obtain the transition states at different times t. :

[0057] (1)

[0058] in, It is a hyperparameter that monotonically increases with time step t and follows a certain condition. and ; It is a hyperparameter that controls the noise variance; It is the identity matrix. Analysis of equation (1) reveals that when the number of noise addition steps is T, the initial state... ,in .

[0059] S202, in order to iteratively start from the initial state Sampling yields the restored image ResShift employs the following random sampling process:

[0060] (2)

[0061] in, It is obtained by reparameterization from a deep network; when hour, By using the initial state Substituting into equation (2) and iterating for T steps, we will eventually obtain a random restoration result. .

[0062] S203, in order to solve the random sampling problem in equation (2), it was modified to obtain the following deterministic sampling process for ResShift as the teacher network:

[0063] (3)

[0064] in, It is a unit impulse function; It is the restored image predicted from the pre-trained random ResShift. Parameters , and They are as follows:

[0065] (4)

[0066] Unlike traditional diffusion models based on random sampling used as teacher networks, this invention improves upon the original random sampling diffusion model, deriving and implementing its deterministic version. Using a deterministic sampling-based diffusion model as a teacher network offers superior characteristics. Specifically, this method significantly improves the stability and consistency of the inference process, reduces the randomness of the generated results, and thus enhances the model's controllability over input conditions while ensuring generation quality. Furthermore, deterministic sampling reduces fluctuations in results during repeated experiments, improving the reproducibility and reliability of experiments, making it more advantageous in practical applications. By introducing deterministic sampling, the model responds more accurately to input conditions, enhancing its control over details and overall structure in generation tasks, especially in scenarios requiring high-fidelity reconstruction (such as MRI image super-resolution).

[0067] Specifically, the process of S102 includes the following steps:

[0068] S301 proposes utilizing student networks. To learn stochastic initial states from a teacher network Compared with deterministic output A deterministic mapping between them. Specifically, the following distillation loss was adopted. Constrained initial state Single-step output in a student network Compared with deterministic output The degree of approximation between them:

[0069] (5)

[0070] in, It is a student network that can directly predict the restored image in just one step; It is a proposed deterministic process regarding ResShift; This represents the Mean Squared Error (MSE) loss function.

[0071] S302, considering that the distillation strategy shown in equation (5) does not use real images during training. This limits the performance ceiling of the student model. To further improve the performance of the student model, a method combining real images is proposed. Consistency loss To provide additional regularization constraints:

[0072] (6)

[0073] S303, Ultimately, the overall training objectives for the student network are as follows:

[0074] (7)

[0075] The design of the loss function for the student network in this embodiment of the invention includes two parts: loss based on distillation strategy. And the loss of consistency with the real image . Teacher network output with deterministic sampling As the learning objective of the student network, minimizing the student network... Knowledge distillation is achieved by analyzing the error between the network output and the teacher's output. This method differs from traditional random sampling distillation, offering greater stability and consistency. Furthermore, it incorporates data from real images. Consistency loss This approach compensates for the limitation of the distillation strategy in not directly using real images during training, thereby improving the student network's ability to represent target images. The distillation loss is optimized from the guidance information provided by the teacher network, while the consistency loss directly constrains the difference between the generated results and the real images. This dual constraint improves the model's learning efficiency and generation quality.

[0076] Specifically, such as Figure 3As shown, the student network adopts the Unet network architecture, which consists of a feature extraction path and a feature fusion path. The feature extraction path mainly includes three stages: feature compression, feature reconstruction, and final output. The feature compression stage includes shallow, intermediate, and deep modules to extract progressively more detailed feature representations. Each module consists of convolution operations, cross-attention mechanisms, and downsampling operations. The feature reconstruction stage also includes deep, intermediate, and shallow modules to gradually restore feature details. The deep module consists of convolution, cross-attention, and normalization operations, while the intermediate and shallow modules use convolution, cross-attention, and upsampling operations for restoration. The final output stage generates the restored image at step T through convolution, cross-attention, and upsampling operations. The feature fusion path is responsible for matching and fusing features from different levels in the feature extraction path. For example, it aligns deep, intermediate, and shallow features with the corresponding reconstruction module features before inputting them, thereby enhancing the reconstruction effect. Furthermore, a cross-attention mechanism runs throughout the entire network, integrating the time step T with the low-resolution image... Information is embedded in modules at each stage to enhance the network's expressive power.

[0077] The Unet network architecture, through its layered design of deep, medium, and shallow modules, effectively extracts and retains multi-level feature information, ensuring a balanced capture of details and semantic information and improving the overall quality of image restoration. The refined feature matching strategy in the feature fusion path enhances the reconstruction module's ability to express target features, thereby improving the detail restoration effect of the restoration results. The cross-attention mechanism runs through the entire network, enabling the model to better capture global contextual information, especially performing better in complex scenes or multimodal inputs (such as time steps and low-resolution images).

[0078] Specifically, experiments were conducted on embodiments of the present invention. The performance metrics of embodiments of the present invention were compared with those of other super-resolution methods, as shown in the table below:

[0079]

[0080] Experimental results show that the proposed method performs excellently in terms of PSNR (24.90) and SSIM (0.675), fully demonstrating its advantages in image quality and structural information preservation. Furthermore, the LPIPS value of 0.221 is the lowest among all compared methods, further indicating its significant advantage in visual quality. Overall, the single-step diffusion-based MRI image super-resolution method proposed in this application demonstrates superior performance in image detail restoration and visual effect enhancement.

[0081] See Figure 4 The diagram shown is a system structure diagram according to an embodiment of the present invention, including:

[0082] Teacher network construction module 401 constructs a conditional diffusion model based on deterministic sampling as the teacher network;

[0083] Student network construction module 402 constructs a student network based on distillation loss and consistency loss, based on the teacher network.

[0084] Image super-resolution module 403 utilizes a student network to achieve MRI image super-resolution.

[0085] See Figure 5 The diagram shown is a schematic representation of the hardware structure of an electronic device provided in an embodiment of the present invention. Figure 5 As shown, the electronic device of this embodiment includes a processor 501 and a memory 502; wherein the memory 502 is used to store computer execution instructions; and the processor 501 is used to execute the computer execution instructions stored in the memory to implement the various steps performed by the electronic device in the above embodiment. For details, please refer to the relevant descriptions in the foregoing method embodiments.

[0086] Alternatively, the memory 502 can be either standalone or integrated with the processor 501.

[0087] When the memory 502 is set up independently, the electronic device also includes a bus 503 for connecting the memory 502 and the processor 501.

[0088] This invention also provides a computer storage medium storing computer execution instructions, which, when executed by a processor, implement the method described above.

[0089] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0090] In the embodiments provided by this invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0091] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0092] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0093] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0094] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0095] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.

[0096] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0097] The aforementioned storage medium can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium accessible to general-purpose or special-purpose computers.

[0098] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. Both the processor and the storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic device or host device.

[0099] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0100] As can be seen, this invention effectively extracts key information from the model generation path by redesigning the diffusion model training process, enabling the model to transform from a complex multi-step generation process into a one-step direct generation of high-resolution images. The distillation strategy of this invention condenses the potential distribution of the diffusion model into a single-step generation by progressively reducing redundant steps in the diffusion process, while fully preserving the model's reconstruction capabilities and image quality advantages. Compared with existing fast solution methods, this strategy significantly improves inference efficiency, not only reducing inference time but also maintaining or even surpassing the quality of traditional methods in the generated images. This distillation-based single-step generation method not only solves the computational bottleneck of existing methods but also provides a new research direction for the further development of diffusion models.

[0101] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A super-resolution method for MRI images based on single-step diffusion, characterized in that, Includes the following steps: A conditional diffusion model based on deterministic sampling is constructed as the teacher network; Construct a student network based on distillation loss and consistency loss using a teacher network; Super-resolution of MRI images using student networks; Specifically, the construction of a conditional diffusion model based on deterministic sampling as a teacher network involves using a conditional diffusion model based on random sampling as the base model, replacing the random sampling in its generation process with deterministic sampling, and thus obtaining the teacher network. The construction of the teacher network based on deterministic sampling includes the following steps: A pre-trained conditional diffusion model, ResShift, is introduced, which uses random sampling from the initial state. Sampling yields the restored image ; By modifying the random sampling of ResShift, a teacher network based on deterministic sampling is obtained; the deterministic sampling process is represented as follows: ； ； in, This indicates a random sampling process. This represents the state at step t. Represents low-resolution MRI images. This represents a hyperparameter that monotonically increases with time step t. It is a unit impulse function; It is the restored image predicted from the pre-trained random ResShift; where Indicates the noise correction term. The offset parameter represents the deterministic sampling. This represents the weighting coefficients that change with the time step. The construction of the student network based on distillation loss and consistency loss includes the following steps: The Unet network architecture is used as the foundation network for the student network; Constructing distillation loss Constrained initial state Single-step output in a student network Compared with deterministic output The degree of approximation between them is expressed as: ； in, It is a student network that can directly predict the restored image in just one step; It is a deterministic single-step generating function of ResShift; This represents the mean squared error loss function; Constructing realistic images Consistency loss , is represented as: ； The overall training objective for constructing the student network is expressed as: ； in, Indicates the training objective. This indicates finding the parameter corresponding to the minimum value. Represents the expectation operator based on multiple variables; The basic network of the student network is trained using the overall training objective, resulting in a student network based on distillation loss and consistency loss.

2. The MRI image super-resolution method based on single-step diffusion according to claim 1, characterized in that, The student network includes: The feature compression stage includes a shallow compression module, an intermediate compression module, and a deep compression module connected in sequence. These modules extract shallow, intermediate, and deep compressed feature maps of the input noisy image through convolution and downsampling, respectively. The feature reconstruction stage includes a deep reconstruction module, an intermediate reconstruction module, and a shallow reconstruction module connected in sequence. The deep reconstruction module receives the deep compressed feature map and obtains the deep reconstructed feature map using convolution and normalization layers. The intermediate reconstruction module receives the deep reconstructed feature map and the deep compressed feature map transmitted through the feature fusion path, and obtains the intermediate reconstructed feature map using convolution and upsampling. The shallow reconstruction module receives the intermediate reconstructed feature map and the intermediate compressed feature map transmitted through the feature fusion path, and obtains the shallow reconstructed feature map using convolution and upsampling. In the final output stage, the system receives the shallow reconstructed feature map and the shallow compressed feature map transmitted through the feature fusion path, and uses convolution and upsampling to generate the restored image at step T. .

3. The MRI image super-resolution method based on single-step diffusion according to claim 2, characterized in that, The shallow compression module, intermediate compression module, and deep compression module all include convolutional layers, cross-attention, and downsampling units with a stride of 2 connected in sequence; The deep reconstruction module includes convolutional units, cross-attention layers, and normalization layers. The intermediate and shallow reconstruction modules each include convolutional layers, cross-attention layers, and upsampling units with a stride of 2 connected in sequence. The final output stage consists of sequentially connected convolutional layers, cross-attention, and upsampling units with a stride of 2.

4. A single-step diffusion-based MRI image super-resolution system, used to implement the single-step diffusion-based MRI image super-resolution method according to any one of claims 1 to 3, characterized in that, include: The teacher network construction module builds a conditional diffusion model based on deterministic sampling as the teacher network; The student network construction module builds a student network based on distillation loss and consistency loss, building upon the teacher network. The image super-resolution module utilizes a student network to achieve MRI image super-resolution.

5. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-3.

7. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-3.

Images